You are here:
A Systematic Evaluation of Analogs and Automated Read-across Prediction of Estrogenicity (QSAR2016)
Citation:
Pradeep, P., K. Mansouri, G. Patlewicz, AND R. Judson. A Systematic Evaluation of Analogs and Automated Read-across Prediction of Estrogenicity (QSAR2016). Presented at QSAR 2016, Miami, FL, June 13 - 17, 2016. https://doi.org/10.23645/epacomptox.5189395
Impact/Purpose:
slide presentation at QSAR2016 meeting on Systematic Evaluation of Analogs and Automated Read-across Prediction of Estrogenicity
Description:
Read-across is a data gap filling technique widely used within category and analog approaches to predict a biological property for a data-poor (target) chemical using known information from similar (source analog) chemical(s). Potential source analogs are typically identified based on structural similarity. Although much guidance has been published for read-across, practical principles for the identification and evaluation of the scientific validity of source analogs remains lacking. This case study explores how well 3 structure descriptor sets (Pubchem, Chemotyper and MoSS) are able to identify analogs for read-across and predict Estrogen Receptor (ER) binding activity for a specific class of chemicals: hindered phenols. For each target chemical, analogs were selected using each descriptor set with two cut-offs: (1) Minimum Tanimoto similarity (range 0.1 - 0.9), and (2) Closest N analogs (range 1 - 10). Each target-analog pair was then evaluated for its agreement with measured ER binding and agonism. The analogs were subsequently filtered using: (1) physchem properties (LogKow & Molecular Volume), and (2) number of literature sources as a marker for the quality of the experimental data. A majority vote prediction was made for each target phenol by reading-across from the closest N analogs. The data set comprised 462 hindered phenols and 257 non-hindered phenols. The results demonstrate that: (1) The concordance in ER activity rises with increasing similarity, (2) none of the 3 descriptor sets are clearly superior to the others for analog identification for ER read-across, (3) selecting hindered versus non-hindered phenols as analogs does not significantly improve concordance in ER activity, and (4) filtering analogs using physchem properties improves overall concordance. The read-across predictions reveal that: (1) PubChem and Chemotyper descriptors are superior to MoSS for ER activity, and (2) filtering of analogs significantly increases the prediction accuracy. As an example, the prediction accuracy using 3 closest analogs from PubChem is 70%. This increases to 74% when filtering by physchem properties, and 89% when data quality is accounted for. This case study demonstrates how biologically-relevant chemical descriptors can be used to identify valid analogs for read-across. This abstract does not necessarily reflect U.S. EPA policy.
URLs/Downloads:
DOI: A Systematic Evaluation of Analogs and Automated Read-across Prediction of Estrogenicity (QSAR2016)
QSAR2016_POSTER_PPRADEEP.PDF (PDF, NA pp, 480.481 KB, about PDF)
PPRADEEP_QSAR2016-ABSTRACT_GP_RSJ.PDF (PDF, NA pp, 94.651 KB, about PDF)